 "But it can be useful, like for summaries!" Meanwhile, 99.9% of generative AI usage is for replacing authors and editors with garbage writing, outright scams, flooding search engines with shit, ripping off artists, cheating on writing assignments, firing translators, creating pornographic images of people who didn't consent, generating CSAM, stealing other people's code and advancing the fine art of low effort enshittification. 54 comments
   There is a Dutch saying, “convenience serves people” (gemak dient de mens), but I feel it always ends up the other way around.🙃 @thomasfuchs THIS RIGHT HERE. @gwynnion @thomasfuchs Was the earlier software you’re referring to some kind of LLM? Or was there a different kind of process that could generate summaries? @stephenjbell @gwynnion @thomasfuchs Different process. Among other things, NeXTSTEP had summarization of blocks of text as a system-level service over 25 years ago. It worked on a 25 MHz Motorola 68040. The functionality lives on in macOS, since it directly descends from NeXTSTEP. With some text selected, Apple menu > Services > Summarize. @thomasfuchs @stephenjbell @gwynnion I wasn’t sure sure if they had it in the 80s, but I knew they had it in the 90s a few years before Apple bought them. @thomasfuchs the quote is one of the many arguments for prompt-to-text generators that doesn't make sense to me. How does the prompter know it's a good summary if they don't know how to make summaries? Summarizing a story is also one of the most efficient ways for a writer to understand what they wrote. The actual summary doesn't do that as much as the process of condensing passages does. They're missing that process. That brings me back to thinking they don't know if the result is any good. @thomasfuchs and that reminds me of the machine translations for fiction. The people putting these out don't recognize how bad the translations are, yet they insist they are translated well because the page at a glance might look like decent prose. That's not as useful as its fans insist it is. Except for scams. Just this week I went to the net to see if there was a practical differentiation between "Sword" and "Brand" as a little used synonym: did "Brand" apply to specific sword makes or styles, or was it an etymological crossover. The obviously AI Generated response assured me this was a common question, and to improve my writing I should know a *Sword* is a weapon and a *Brand* is product identity. Yeah, I trust AI translations. Riiiight. @shadowsminder @thomasfuchs It depends how often you need "good" summaries. The reddit bot that summarized articles in the comments was "good enough" the majority of the time. And people reading the comments liked it better than having no summary at all, which was the alternative. To me this is similar to auto generated captions, which are noticeably worse than the summaries. But they're still so much better than no captions at all. Humans would do it better, but they weren't doing it. @thomasfuchs Out of the corner of my eye I'm watching a presentation from a company and their data is saying that it's mostly used to replace search and I think that a) that's a fucking indictment of search companies b) pretty naive of them to believe that's what it's being used for mostly.  @thomasfuchs @thomasfuchs well, at least AI cannot replace physical labor, you know illegal immigrants can do that./s According to the people working to divide US. Though AI is world wide. @Leefellerguy @thomasfuchs https://newatlas.com/robotics/figure-01-openai-humanoid-robot-real-time-conversations @thomasfuchs I keep thinking about how much energy is being used to do those things with "AI"... I still refuse to flat-out call it AI since it isn't intelligent lol @thomasfuchs If you want to see how "well" AI can summarize, @Jess_Alter did a whole piece on it. It summarized one of her stories so poorly it was laughable. It basically started with a goodreads existing piece and then went off the rails. https://ko-fi.com/post/Interview-with-a-Data-Vampire-N4N2UU0VI  @thomasfuchs @thomasfuchs @thomasfuchs there is an old way of words that has returned. A.i. did it's best but it is just a compilation of human minds, the data it has is old and needs updated. That's where things like old druid poetry and words from the afterlife come into play. Humans have done it for generations and the only thing a.i. lacks is the feeling behind it's words. Don't forget about the mind-boggling use of writing scientific paper's abstracts. You can save an hour of work all the while undermining the legitimacy of the actual paper (which most likely took years of work, assuming is not %100 LLM hallucinations though). @excess @thomasfuchs https://defector.com/the-brief-and-wondrous-life-of-the-ai-giant-penised-rat-explained  @thomasfuchs they will use it to back every job they can. I think Amazon will be the first try to replace everyone with A.I. including physical jobs... @thomasfuchs For me, the funny point is that those people usually disregard that exact same argument for crypto currencies/NFT. I personally have no use for either and most usages are worthless, but there _are_ legitimate use cases for both. @thomasfuchs Google for help in any common topic and that should be enough to demonstrate just how much AI generated content has already ruined the web. So many articles that are full of right keywords, but then it's just page after page of unhelpful and worthless junk, being used to funnel people into ad pages.  @thomasfuchs What worries me about generative models making summaries is their tendency to „average out“ ideas that are outside the norm. I’ve experienced that on brainstorming, trying to expand cliffnotes into text, and for that same reason I do *not* trust these models with preserving exceptional, new ideas that don’t align with the mainstream. It’s a great way to poison your potential for new insights.   @thomasfuchs Also maybe not so much on summaries either. https://bymattruff.com/2024/03/26/in-which-an-ai-hallucinates-the-content-of-my-novels/  Generally true except I don't see how it steals other people's code, I suspect its trained on open source repos anyway and doubt it reproduces entire large programs without serious NYT-level prompt engineering Also it floods the internet with shit, not the search engines. The search engines (looking at you google) are responsible for promoting this shit in their algorithms @thomasfuchs Fair but I have yet to see reports of AI replicating large chunks of code wholesale @thomasfuchs A.I. is the problem we create to deal with all the other problems we created, not unlike introducing a new species with no natural predators to control the previously introduced species that had no natural predators, etc. @thomasfuchs Ok, but… citation needed? Nobody can deny that AI is generating tons of junk, but how are you judging the amount of legit and useful applications? Those don’t get dumped out into the wild as often, so we don’t know about it.  @thomasfuchs I'm constantly flummoxed by how "generating summaries" is considered useful work. Like, are there *really* so many people out there who are required to summarize "other people's stuff" but *also* don't need to learn the stuff themselves in order to summarize it? > “Yay, I can stop reading stuff I don't need to and can just clickity-clack to get generated summaries of it all instead!” **Said me never.** Who's doing this??!!?! Managers, project managers and execs. There's a problem with building things in a giant hierarchy: it's really hard to tell what's going on. Depending how you try to solve it you get silos, you get fierce code ownership, you get confused code touched by too many hands, you get duplicated effort. Not clear how you fix that. But "propagating summaries of work done to everybody, and being able to summarise at many levels of detail" sounds like a possible answer, right? Not saying LLMs will fix that. I'm pretty sure they won't. But that's part of why "summarising" sounds so good to so many people. If you're overseeing a big chunk of people, you *know* you don't know what's going on very well and you wish you had somebody to read and summarise all the stuff going on inside it. Of course, if we had good summaries of this stuff we'd discover how bad the quality of the text being summarised was. One problem at a time? @jaredwhite @thomasfuchs But "provide a view and summary of a lot of other people's stuff without understanding it" is kind of the *definition* of middle management, as viewed from the executive side. You might ask, "wait, management wants to replace *themselves* with LLMs?" And I'd say "no, but execs do." Managers are very expensive. Managers are aware that *good* managers do a lot more than that. But that's kind of like engineers understanding LLMs can't do their job - doesn't stop execs. @codefolio @thomasfuchs But now we're back to "I don't understand their jobs, so they must be expendable." That's turtles all the way down…or up as the case may be. Perhaps all a CEO does is get "summaries" from the division heads / partners / etc. and then calculates the best option out of a series of possibilities when planning for the future. Hmm, can AI do that? (Yes! Says anyone *not* a CEO.) I do get what you're saying. Maybe "bad" managers can be swapped out. But that's not saying much… Well, and if you pretend you had an excellent free/cheap summariser, there might be a lot you could do with it. That service *would* have applications. A lot of businesses put a *ton* of people-time into trying to get the same effect by making people send out various levels and types of status updates in all directions. An LLM won't do that well, but there *is* a business desire for a thing that would. @thomasfuchs It is not AI. That's the key point. It means to be fucking with our language and our brains to call LLMs and SDs Artificial Intelligence. They are static networks of vector nodes, products of mathematic iteration, incapable of self reflection and adaptation. They are bayesian token chainguns. But they are not intelligent. |
Gregory's Server
@thomasfuchs this, it's shocking how many people just completely ignore the negative externalities of the little bit of convenience they gain.